was analyzed separately for the two classes. For each one of the FA1 line, the coefficient of variation
abbreviated as CV was calculated for each one of the nine characters. Variability within FA1 lines is
generally greater that variability of the FC popula- tion Table 6. Moreover, significant differences
may be observed in the level of variability between the two classes. Variability is considerably in-
creased in FA1 lines of Class II for the number of seeds or the relative stem height, but the situation
is reversed for average seed weight Table 6. Accordingly, lineages belonging to Classes I and II
cannot be simply distinguished by this analysis. The CV values of the two FA1 lines from the same
lineage were plotted together. No relationship is observed in Class I P \ 0.05, Fig. 4A, while a
significant positive correlation r = 0.50, 25 d.f., P B 0.01 appears in Class II Fig. 4B.
A capacity for developmental reversion was especially observed in Class II lineages Fig. 3,
which also showed a ‘control’ of variability Fig. 4B. The link between these two types of events was
tested. For each FA1 line and for each character, the coefficient of variation was standardized as a
function of the average CV value of the character calculated for the FC population see Table 5.
These standardized values abbreviated as SCV were calculated as described above for mean values
SV, see Section 2. Then, for each character and for each FA1 line, the standardized coefficient of
variation SCV was plotted as a function of the standardized mean value SV. No significant rela-
tionship P \ 0.05 appears for FA1 lines of Class I Fig. 5A, while a negative significant correlation
P B 0.01 is observed for FA1 lines of Class II Fig. 5B.
4. Discussion
4
.
1
. Methodological considerations An a posteriori distinction between two classes
of lineages is problematic, because the same crite- rion cannot separate populations and be used to
derive conclusions regarding their differences. Without any independent criterion of separation of
the classes, the discrepancy from standard values observed in FA0 but not in FA1 of Class II may
Table 6 Coefficient of variation of FA1 lines from Class I and Class II
a
LDW SDW
SH TSW
NS ASW
SH:SDW ratio NS:SDW ratio
TSW:LDW ratio Class I FA
1
lines 37.4
29.0 4.55
40.9 530
1
35.3 16.6
42.1 20.3
19.3 530
2
33.9 39.5
9.25 53.5
44.8 18.0
33.6 39.3
43.0 46.0
715
1
33.0 25.8
31.6 23.5
23.3 40.4
24.5 6.82
26.0 715
2
13.6 23.9
17.5 20.6
14.3 18.0
27.8 3.55
40.7 34.6
54.6 22.4
38.8 716
1
45.3 4.49
34.3 35.5
25.0 4.93
29.0 22.9
27.0 28.8
34.1 28.2
28.3 716
2
Class II FA
1
lines 46.0
37.7 14.0
42.9 46.9
17.21 24.0
21.1 44.7
611
1
67.9 69.9
6.62 22.0
7.1 25.2
611
2
27.5 58.3
56.7 51.9
44.0 14.7
50.5 57.9
701
1
59.5 42.2
39.2 6.97
33.3 24.9
8.92 76.1
701
2
73.8 9.7
36.3 67.2
60.5 19.9
33.7 6.76
34.6 718
1
39.5 18.9
51.9 29.2
28.2 35.3
35.0 4.33
47.2 718
2
41.8 17.3
52.7 22.0
39.5 A6erage of CV 6alues
Class I 30.0 a
5.60 a 40.1 a
31.4 a 22.2 a
33.9 ab 27.6 a
31.6a 30.8 a
29.5 a 29.8 a
8.47 a Class II
48.2 b 54.4 b
46.8 b 42.8 a
13.6 b 51.7 b
5.94 a 28.6 a
28.1 a FC lines
28.5 a 30.2 b
15.7 b 34.0 a
30.8 a 40.4 a
a
The same letter following two values in a column indicates that they are not significantly PB0.05 distinguishable by a t-test. Abbreviations as in Table 1.
Fig. 4. Relationship between variability within two FA1 lines from the same lineage. A Class I lines, B Class II lines. For
each parameter, the standardized values of coefficient of varia- tion SCV, see Section 3 of a FA1 line is plotted as a function
of the corresponding value for the other FA1 lines from the same lineage. No significant correlation is observed in 4A
r = 0.0098, 25 d.f., while a positive significant correlation P B 0.01 appears in 4B r = 0.50, 25 d.f..
parameters is not verified, because salt-adaptation generates
plurimodal distributions
Amzallag, 1998; ii A third source of variability has been
identified during the process of salt-adaptation. This source is neither genetic nor environmental
Amzallag, 1999a, but is generated by self-orga- nized processes involved in critical phases in de-
velopment
Amzallag, 2000a,b.
Accordingly, analysis of the variability cannot be analyzed by
ANOVA, nor can it be analyzed by any other method requiring a definition of the groups before
the experiment. A posteriori segregation of the initially homogeneous population remains the
only way to analyze the phenomenon of ‘develop- mental reversion’. Moreover, it is likely that phe-
nomena
of developmental
reversion are
so frequently ignored because of the methodological
limitations of analyses based on an a priori segregation.
Fig. 5. Relationship between deviation in mean and CV values in FA1 lines. A Class I lines, B Class II lines. For each
parameter and for each FA1 line, the standardized mean value SV is plotted as a function of the standardized CV value
SCV, see Section 3. No significant correlation P \ 0.05 is observed in 5A r = 0.031, 52 d.f., while a negative signifi-
cant P B 0.01 correlation is observed in 5B r = − 0.357, 52 d.f..
be explained as a random fluctuation in some FA0 plants, eventually due to an artifactual mod-
ification of the environment. However, the ‘ran- dom fluctuation hypothesis’ cannot explain both
the differences in variability between the two classes Fig. 4 and the link between perturbation
and variability Fig. 5. The range of expression and variability of the characters is similar for the
two classes, but the link is observed only in Class II. Consequently, analysis of variability provides
the validation for methodological segregation be- tween the two classes of lineages in the initially
homogeneous population.
Tests of analysis of variance ANOVA are generally used in order to separate the genetic and
environmental sources of variation. Beyond the methodological problems related to interpretation
of ANOVA see Lewontin, 1974, this method is not adequate in the present case for two reasons:
i The hypothesis of normal distribution of the
4
.
2
. The phenomenon of canalization Offspring of salt-adapted plants are modified in
their reproductive development Table 1. How- ever, even in the absence of selection see Section
2, modifications are reduced after two genera- tions of salt-adaptation treatment Table 2. The
fact that FA1 plants developed in a non-salinized medium is not sufficient to explain this phe-
nomenon because deviation occurred specifically in some of the FA1 lines Class II lineages, Table
4. Moreover, deviation is larger for FA0 than for FA1 plants Table 2 although they were grown in
the same environment at the same time see Sec- tion 2. These considerations confirm that salt-
adaptation induces modifications in reproductive development Amzallag, 1998, which are able to
influence the next generation Amzallag, 1996; Amzallag et al., 1998. They reveal a capacity to
recover an almost-normal development after more than one generation of salt-adaptation.
A similar recovery of perturbations has been observed in plants regenerated from cultured cells
following two sexual generations Oono, 1985; Morrish et al., 1990; Cheng et al., 1992. In
general, such a phenomenon has been interpreted as a reversion, assuming that the modifications
were caused by reversible genetic changes such as changes in the pattern of cytosine methylation
leaving the DNA sequence unaffected. Accord- ingly, the phenomenon of reversion is interpreted
as an elimination of the stress-induced modifica- tion following re-exposure to the optimal condi-
tions. Such an interpretation cannot be rejected, but is not likely in the present case, because
developmental reversion also occur for plants maintained in salinity for many generations
Amzallag and Seligmann, 1998. Moreover, a capacity to recover an almost-initial status was
clearly demonstrated after occurrence of non-re- versible genetic changes.
Working with maize mutants, Martienssen et al. 1990 described a non-random modification in
the genome structure, during the early vegetative development, which enabled the seedlings to re-
cover a wild-type phenotype. They suggested that a genetic mechanism is able to restore a functional
morphogenetic capacity, in spite of non-reversible modifications accumulated during previous gener-
ations. Such a phenomenon also occur during normal development. For example, the cellular
DNA content of embryos of Dasypirum 6illosum Graminae varies relatively to position of the
caryopse on the spike Frediani et al., 1994. This variation is not due to a change in polyploidy, but
rather to amplification andor deletion of specific repetitive
DNA sequences.
Nevertheless, the
DNA content is buffered towards a ‘standard value’ during the early development of the
progeny of low or high DNA content. This pro- cess is not random: repetitive DNA sequences
which were the most deleted are preferentially amplified during this self-regulation process Fre-
diani et al., 1994. These examples indicate that a property of auto-regulation exists at the genome
level despite large non-reversible modifications see Amzallag, 1999b and references therein.
Canalization, as defined by Waddington 1957, does not imply any reversibility of the changes
involved in expression of the characters temporar- ily modified. For this reason, canalization is more
appropriate
than re6ersion
to describe
this phenomenon.
4
.
3
. Mechanisms of canalization The developmental perturbation was not al-
ways higher in Class II plants than in Class I ones Table 5. However, at a similar level of deviation,
Class II plants showed canalization while Class I plants did not Fig. 3. This suggests that canal-
ization is not expressed independently for each parameter, but is triggered for the reproductive
development as a whole, in response to a global level of perturbation.
In progeny of salt-adapted Sorghum, the recov- ery of a normal reproductive development was
related to an increase in connectedness between reproductive
characters, resulting
from their
isolation from the modified vegetative organs Amzallag and Seligmann, 1998. Such compart-
mentation of the development has been consid- ered as a mechanism for increasing developmental
stability in presence of genetic or environmental variations
Conrad, 1990;
Amzallag, 2000a.
This process
of strengthening
by isolation
explains why canalization may be observed for the reproductive development as a whole, and not
only for the characters with the most perturbation Fig. 3.
The recovery of reproductive development through isolation from the vegetative organs is
expected only in the case of informational redun- dancy between, and within, successive phases of
development. Such redundancy of the information was recognized as a factor in the stability of
development Conrad, 1990; Chauvet, 1993, and it was experimentally observed during vegetative
development in S. bicolor Amzallag, 1999c. Thus, it is likely that canalization in reproductive
development is reached by elimination of the dis- turbing element the linkage with the previous
phase of development modified by salt-adapta- tion of this redundant informative system.
A negative correlation between level of variabil- ity and mean value respectively, SCV and SV
was observed only for FA1 lines of Class II Fig. 5. This indicates that low variability is not di-
rectly related to expression of the standard values Fig. 5A, but that it is especially reduced by the
process of canalization itself Fig. 5B. In other words, the decrease in variability is linked to
recovery of an optimal value. This suggests that optimal values act as ‘thermodynamic basins of
attraction’ see Conrad 1990 for discussion of this notion for the process of canalization. Ac-
cordingly, canalization becomes an expression of the spontaneous tendency of living open systems
to reach their minimum free-energy level Pri- gogine and Wiame, 1946; von Bertalanffy, 1950.
5. Conclusions
